Violating closure and detecting trap-shyness

[MWhitehead]

Hi all,

I have been carrying out mark-recapture on a population of wasps. I have been marking them individually, and every day I record recaptures and mark unmarked captures. I have estimated abundance using the Poisson-log normal mark-resight model for robust design. This is all great and I have numbers that make some biological sense.

Now I would like to test whether or not wasps become "trap-shy". So for this I have been thinking regular old closed captures models should give me an impression of how p (capture probability) varies with time throughout the study, and it appears time-varying p models best fit the data. My hypotheses then would be:

H1: Wasps are becoming trap shy: estimated p (capture probability) should drop with time, as the proportion of marked (previously trapped animals) increases in the population.

H2: If wasps do not become trap shy, then p should remain constant or increase as the proportion of marked individuals in the pop increases.

The only trouble is that I might be violating closure.

My wasps don't move much. The median movement from day to day is an order of magnitude less than the scale of my sampling (<20m median, 300m x 100m long plot), however I must admit the possibility of some immigration/emigration in and out of the site.

My questions are this:

a) How would some violation of closure affect my estimates of capture probability? If anything, I assume that estimated capture probabilities should be downwardly biased if marked wasps are leaving or unmarked ones entering.

b) Does this seem like a reasonable way to examine "trap-shy" behaviour? Are there more refined ways to test this particular hypothesis? Any input welcome.

I must say, after a very steep initial learning curve it is possible for MARK analysis to become fun!

Thanks for any help.

[murray.efford]

As you've not drawn another reply, here is my 2c worth:
1. Why use mark-resight? From your description it sounds like you have mark-recapture data.
2. The usual way to investigate trap shyness would be to compare models with and without this effect, rather than try to infer it from trends in capture probability over time.
3. My simulations in 'secr' suggest roughly 25-30% of the wasps you catch could have 'home-range' centres outside the plot (assuming they have something like a home range). This may not be ignorable. (% depends on absolute detection probability and number of sampling occasions).
4. Yes, temporary absence from the plot lowers the estimate of detection probability, although it's moot whether this should be called 'bias'.
5. If these are solitary wasps with something like a home range, and you recorded capture locations, then a spatial analysis may be appropriate (see e.g., Efford in press Ecology). This would allow you to compare models with and without trap shyness while allowing for movement across the plot boundary.
Murray

[MWhitehead]

Thank you so much for the helpful reply Murray. I only wish I had seen it earlier. I think the initial quiet response and going O/S shortly after I made this thread caused my question to slip my mind!

In answer to your question:
I was a little unclear in my initial post. I have made things difficult for myself by employing two sampling strategies. The first series of occasions (n=9) are done at 9 fixed points in the method I first described. The latter series of occasions (n=9) are done randomly around the site with *no marking*, just recording recaptures. Therefore, I used a mark-resight model to estimate pop densities as it could handle the unmarked capture data I collected in the latter trapping series.

As the former trapping series is done intensely at only 9 fixed trap points, I felt that the data for this series would be ideally suited to testing for trap-shyness at the fixed traps.


Your suggestion to carry out a spatial model is very helpful to me. I've read your paper now and the documentation for secr and the polygon detector spatial analysis looks like a perfect fit for the data I have. I will try a g0 ~ b model and see how that looks.

Thanks again for the help. Now to wrestle with R.